X-Original-To: pgsql-admin-postgresql.org@localhost.postgresql.org Received: from localhost (unknown [200.46.204.144]) by svr1.postgresql.org (Postfix) with ESMTP id D78315289D for ; Fri, 17 Jun 2005 18:37:55 -0300 (ADT) Received: from svr1.postgresql.org ([200.46.204.71]) by localhost (av.hub.org [200.46.204.144]) (amavisd-new, port 10024) with ESMTP id 28276-10 for ; Fri, 17 Jun 2005 21:37:54 +0000 (GMT) Received: from cmailm3.svr.pol.co.uk (cmailm3.svr.pol.co.uk [195.92.193.19]) by svr1.postgresql.org (Postfix) with ESMTP id 1864352896 for ; Fri, 17 Jun 2005 18:37:52 -0300 (ADT) Received: from modem-1975.lynx.dialup.pol.co.uk ([217.135.199.183] helo=192.168.0.102) by cmailm3.svr.pol.co.uk with esmtp (Exim 4.41) id 1DjOX3-0006vc-Nk; Fri, 17 Jun 2005 22:37:46 +0100 Subject: Re: Indexes getting corrupted. From: Simon Riggs To: Tom Lane Cc: "Bruno G. Albuquerque" , Scott Marlowe , pgsql-admin@postgresql.org, David Parker In-Reply-To: <14340.1118924727@sss.pgh.pa.us> References: <42A73B60.7010909@dba.com.br> <1118258780.5758.160.camel@state.g2switchworks.com> <42A9C2F7.50306@dba.com.br> <23641.1118422679@sss.pgh.pa.us> <1118912741.3645.206.camel@localhost.localdomain> <14340.1118924727@sss.pgh.pa.us> Content-Type: text/plain Organization: 2nd Quadrant Date: Fri, 17 Jun 2005 22:37:19 +0100 Message-Id: <1119044239.3645.234.camel@localhost.localdomain> Mime-Version: 1.0 X-Mailer: Evolution 2.0.2 (2.0.2-3) Content-Transfer-Encoding: 7bit X-Virus-Scanned: by amavisd-new at hub.org X-Spam-Status: No, hits=0.053 tagged_above=0 required=5 tests=AWL, FORGED_RCVD_HELO X-Spam-Level: X-Archive-Number: 200506/136 X-Sequence-Number: 17891 On Thu, 2005-06-16 at 08:25 -0400, Tom Lane wrote: > Simon Riggs writes: > > I'm suspicious of a more subtle intermittent error. > > Yeah, I am too, but so far none of the reporters have been cooperative > about providing more information :-( > > > We have no > > information about what the magic values are, only that they are not > > correct. Should we increase the information returned for that error? > > I think the proposed patch is a waste of time. What I am hoping to get > from people is a dump of the whole first page of the corrupted index > (via pg_filedump, or even good ol' od). That might give us some idea of > what we are dealing with --- localized corruption in a basically good > metapage, or wholesale replacement of the page with some other page (and > if so what), or maybe it is a hardware fault after all. You can't draw > those sorts of conclusions from one or two words, but with a whole page > to look at you have a shot at telling the difference. Sorry, let me explain. I immediately assumed that all problem reporters had not provided further information because they could not, not because they would not. The patch is trivial and your suggested methods of looking into this are much preferable. However, that relies on somebody being able to locate the metapage, even assuming they don't overwrite the whole thing and continue. The bug is intermittent and so re-running will allow work to continue, so it would seem that is exactly what they do. My patch would allow some more information to be retrieved in the meantime, while we hope for somebody to upload a damaged metapage. If the patch is not sufficient, then I'd suggest that the metapage be dumped to log so it can be more easily provided for inspection. In general, such matters are usually hardware faults. But all reporters of this error in 8.0 have only reported index errors and no others, which leads me to suspect a software related cause. Best Regards, Simon Riggs